| 77 | 0 | 40 |
| 下载次数 | 被引频次 | 阅读次数 |
SlowFast模型存在慢速分支的运算时间过长和对时间特征处理能力不足的问题。针对上述问题，利用FasterNet网络替换Resnet网络作为慢速分支网络，提高慢速分支识别准确率的同时使网络更加轻量化。设计新的预测模块，提高整体网络对时序特征的提取能力。实验表明改进算法的识别准确率在自制数据集上达到85.4%，提升14.3%；在HMDB51数据集上达到87.5%，提升11.3%，运算时间约减少20%。
Abstract：The SlowFast model has problems such as long computation time for slow branches, insufficient processing ability for time features, and inability of the network to meet the accuracy and timeliness requirements for campus abnormal behavior detection. In response to the above issues, it is proposed to use FasterNet network instead of Resnet network as the slow branch network to improve the accuracy of slow branch recognition while making the network more lightweight. Design a new prediction module to improve the overall network's ability to extract temporal features. By conducting ablation experiments and comparative experiments with the improved algorithm model and the initial model, it was demonstrated that the recognition accuracy of the improved algorithm reached 85.4% on the selfmade dataset, an increase of 14.3%; Reached 87.5% on the HMDB51 dataset, an improvement of 11.3%, and reduced computation time by approximately 20%.
[1]丁毅.基于深度学习的多示例学习算法研究[D].天津：天津理工大学，2023.
[2] Feichtenhofer C, Fan H, Malik J,et al.SlowFast Networks for Video Recognition[C]. 2019 IEEE/CVF International Conference on Computer Vision(ICCV). 2019:6201-6210.
[3] Gopalakrishnan T , Wason N , Krishna J R , et al. Comparative Analysis of Fine-Tuning I3D and Slow Fast Networks for Action Recognition in Surveillance Videos[J].Engineering Proceedings,2024,59(1):285-291.
[4] Kim M, Kim T , Kim D.Spatio-Temporal Slowfast Self-Attention Network For Action Recognition[C]. 2020 IEEE International Conference on Image Processing(ICIP).2020:2206-2210.
[5]郭复澳,姚克明,王中洲.基于改进3D-ResNet34模型的视频人员情绪识别研究[J].图像与信号处理, 2024, 13(3):9-14.
[6] Chen J , Kao S H , He H ,et al.Run, Don't Walk:Chasing Higher FLOPS for Faster Neural Networks[C]. 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).2023:21-31.
[7]宋伟,任栋,于京,等.一种新的基于三维卷积共生梯度直方图和多示例学习的特殊视频检测算法[J].计算机学报.2019,(1):149-163.
[8]杨锐,罗兵,郝叶林,等.一种基于深度学习的异常行为识别方法[J].五邑大学学报:自然科学版.2018,(2):23-30.
[9]罗凡波,王平,梁思源,等.基于深度学习与稀疏光流的人群异常行为识别[J].计算机工程.2020,(4):287-293.
[10]聂豪,熊昕,郭原东,等.基于深度学习的视频异常行为识别算法[J].现代电子技术.2020,(24):110-112.
[11]王继坤.基于ResNet的人体异常行为识别方法研究[D].哈尔滨：黑龙江大学, 2022.
[12]窦恩泽,杨鹏程,李小成,等.改进PConvUNet图像修补的解包裹方法[J].现代电子技术.2024,(5):46-52.
[13] Tian Y, Pang G, Chen Y, et al. Weakly-supervised Video Anomaly Detection with Robust Temporal Feature Magnitude Learning[C].2021 IEEE/CVF International Conference on Computer Vision(ICCV),IEEE,2021:4955-4966.
[14] He K, Gkioxari G, Piotr Dollár, et al. Mask R-CNN[C], 2017IEEE International Conference on Computer Vision(ICCV).2017:386-397.
[15] Kuehne H , Jhuang H , Garrote E ,et al.HMDB:A Large Video Database for Human Motion Recognition[J].IEEE, 2011.(5):359-368.
[16]孙健鹏.校园安全数据集[EB/OL].(2024-10-10)[2024-10-11].https://pan.baidu.com/s/10tl7URlJU8gwEQY6AConxA.
[17] Chen J , Wei F , Zhao J ,et al.Revisiting Referring Expression Comprehension Evaluation in the Era of Large Multimodal Models[J].2024.
基本信息:
DOI：
中图分类号:TP18;TP391.41;G434
引用信息:
[1]郭海峰,孙健鹏.基于FasterNet模型的校园异常行为检测[J].通信与信息技术,2025,No.277(05):22-25+34.
基金信息: